Research On Binocular Stereo Vision Measurement Technology

With the introduction of Industry 4.0, intelligent robots with sensory elements have become an important factor in promoting the development of modem industry. Binocular stereoscopic vision is an important method to realize robot automatic measurement, and it is widely used in the navigation and positioning of the robot, the tracking and reconstruction of three-dimensional space, the dynamic measurement of sheet metal forming, the online detection of limit curve and many other fields in modern industry.In recent years, there are still some problems about binocular stereoscopic vision to be improved, and required the researchers to propose some new methods, even though it has made rapid progress. There are mainly concentrated on two aspects: first of all, how to further improve the accuracy of the camera calibration; secondly,, how to balance the relationship between the accuracy and the speed of stereo matching. As a result, this paper focuses on the camera calibration and stereo matching in binocular stereoscopic vision.In this paper, in order to improve the accuracy of camera calibration, the algorithm of the subpixel corner detection was improved in some parts which mainly consists of two steps: First,the corner of the pixel level is obtained by the Harris corner detection algorithm, and then the position of the subpixel angle point is obtained by the method of weighting the gray gradient and the binomial fitting. The experimental results show that the method improves the calibration accuracy of the camera.In order to improve the matching accuracy in binocular stereo vision and reduce the false matching rate in parallax discontinuities and noise interference, in this paper the algorithm based on Census transform was applied. The mainly works are as follows: In the census transformation stage, Replacing the central pixel with the average of its eight neighborhoods to enhance the reliability of the single pixel matching cost; In the cost aggregation stage, in order to accelerate the matching speed, the cost aggregation was done by applied the box filter; In the parallax matching stage, in order to improve the quality of the final disparity map, the proper correlation detection of coarse parallax graph, proper interpolation, subpixel refinement were performed. The experimental results show that the algorithm used in this paper improves the parallax accuracy of low texture and parallax discontinuities, reduces the mis-matching rate and computational complexity, improves the robustness of binocular vision, and maintains good real-time performance.The platform of binocular stereo vision ranging was built, and the experimental results show that the platform built in this paper has a higher range accuracy and a certain application potential.